Research On SOC Estimation And Interpretability Of Lithium Battery For Electric Vehicles

With the establishment of the national carbon peaking action plan before 2030 and the new energy vehicles’ development,lithium-ion batteries will become the important study field of new energy vehicles.The SOC of the battery reflects the remaining capacity of the battery.Accurate battery SOC prediction and its interpretable expression are the prerequisites to ensure battery reliability.For the current traditional lithium battery pack SOC prediction accuracy is not high and more attention is paid Due to the characterization of the information parameters of the battery itself,the influence of the vehicle condition information when the electric vehicle is running on the battery SOC prediction is ignored,and the interpretable expression of the established model is lacking.This paper proposes an improved limit gradient boosting decision tree(DE-GWO-XGBoost)algorithm,and based on this method,an electric vehicle lithium battery SOC prediction model is established.After modeling,the interpretability method is used to analyze the interpretability of the prediction results of the model.The specific work mainly includes:(1)Lithium battery characteristics and modeling analysis.Firstly,taking the battery characteristics as the starting point,analyzing its working principle,comparing and analyzing various battery models and their advantages and disadvantages,and then discussing the characteristics and shortcomings of the current mainstream SOC estimation algorithms,laying the foundation for the subsequent construction and optimization of lithium battery models.(2)Construction and comparison of lithium battery SOC prediction models.Using the actual operation data of electric vehicles,combining the characteristics of different vehicle conditions of electric vehicles with the information representation parameters related to battery safety,a variety of representative machine learning algorithm models are constructed to predict the SOC of lithium battery packs,and from the perspective of evaluation indicators Comparing and analyzing the outcome of each model,the effects indicate that the XGBoost model based on the boosting integration method has lower values of the model evaluation indicators MAE and RMSE,and the fitting effect R2 is better.(3)The proposal and construction of the improved limit gradient boosting decision tree model.Aiming at the problem that the error of the XGBoost model fluctuates greatly,it is further proposed to use the differential evolution algorithm to optimize the model parameters.The effects indicate that the model optimized by the differential evolution algorithm is better on the three evaluation indicators,which verifies the effectiveness of the optimization algorithm.Aiming at the deficiency of the differential evolution algorithm when iterative to a certain area,the difference is reduced and the local optimum appears,the gray wolf optimization algorithm is proposed for fusion,and the reliability index of the model is introduced.Through MAE,RMSE,R2 and model reliability index,the comparative analysis shows that the raised XGBoost model optimized by differential evolution gray wolf algorithm optimization has the best effect among the models compared in this study,which proves the effectiveness of the raised model.(4)The interpretability analysis of the model.Firstly,the importance of the features of the model itself is used to explain,and then the lack of interpretability of the model itself is pointed out.Then,the SHAP theory is used to analyze the interpretability of the improved lithium battery model for electric cars,and the effect of every feature on the prediction result of the model is expounded.And it is the same as the objective facts,which proves the rationality of the improved model in this study,and provides a basis for output decision-making judgment.